The data set comes from the Centre for Research on the Epidemiology of Disasters (CRED). This organisation records every instances of natural disasters since 1900 within the EM-DAT database. This comprehensive open source database complies data from various sources; UN agencies, government agencies, research centers, humanitarian organisations, reinsurance companies and world press agencies. For a full list of sources see the EM-DAT website. I chose to download all the information regarding natural disasters between the year 1922 and 2022 . After looking at the data it was clear that the historic record before 2000 was too sparse to stand up against the quality of data recording conducted by CRED since its inception in 2000. Rather than looking at changes over a century using the historic record I have decided to focus on non historic entries of natural disasters which have occurred since 2000.
To answer my research questions I am interested in where and when different natural disasters occurred. The following variables are of potential interest in asking these questions:
Group - The disaster subgroup:
Type - The specific type of disaster i.e., Drought or Earthquake
Region - Region or continent where the disaster occurred
*Year** - the year the disaster started
Climate Change Effect - whether the natural disaster is a primary direct effect of climate change, secondary indirect effect of climate change or unrelated to climate change. Categorized based on EU report.
for further explanation of each variable see the codebook provided by the EM-DATA database
Has there been a change in prevalence of natural disasters since 2000? Particularly has there been an increase in natural disasters related to climate change i.e. flooding?
What regions are most effected by natural disasters?
# read in data
data <- here("data","emdat.csv") %>% read_csv()
# nrow(data) #16388
# select columns relevant to research question
my_data <- data %>% select(c(Historic, DisNo., Historic, `Classification Key`, `Disaster Group`, `Disaster Subgroup`, `Disaster Type`, `Disaster Subtype`, ISO, Country, Subregion, Region, `Start Year`, `Start Month`, `Start Day`))
#check
# nrow(my_data) #16388# tidy the names of the columns so its in a better format
#changing the names so there are no spaces or capital letter
my_data <- my_data %>% rename(id = DisNo.,
historic = Historic,
classification = `Classification Key`,
group = `Disaster Group`,
subgroup = `Disaster Subgroup`,
type = `Disaster Type`,
subtype = `Disaster Subtype`,
iso = ISO,
country = Country,
subregion = Subregion,
region = Region,
year = `Start Year`,
month = `Start Month`,
day = `Start Day`)
#check the class is correct for every variable
#str(my_data)
#change the class to numeric for year, month and day variable
my_data <- my_data %>% mutate(year = as.numeric(year),
month = as.numeric(month),
day = as.numeric(day))
#look at the data
#total number of disasters per year
kable(my_data %>% group_by(year) %>% summarise(count = n()) %>% head(), caption = "First 6 rows showing historic data. Historic entries before 2000 are markedly lower") %>%
kable_styling()| year | count |
|---|---|
| 1922 | 8 |
| 1923 | 16 |
| 1924 | 9 |
| 1925 | 12 |
| 1926 | 15 |
| 1927 | 10 |
#looks as if the data from before the 90s was markedly lower - it is unlikely due to changes in the environment instead changes in recording quality
#remove any Historic data - data from before 2000 (when the EM-DAT started recording live)
updated_data <- my_data %>% filter(historic == "No")
# nrow(updated_data) #9505
#check i have not lost any data I shouldn't have
my_data %>% nrow() - my_data %>% filter(historic == "Yes") %>% nrow() #9505## [1] 9505Now the data is set up I can create a new column with the climate change information.
#make a new column which categories the type of natural disaster as direct effect of climate change, indirect effect of climate change and not related
# Define vectors of natural disasters classified as primary and secondary effects of climate change
direct_effects <- c("Extreme temperature", "Flood", "Storm")
indirect_effects <- c("Glacial lake outburst flood", "Drought", "Wildfire", "Mass movement (wet)", " Mass movement (dry)")
# Create a new column using a loop which runs through the type column and assigns each entry to the correct climate change condition based on the defined vectors.
full_data <- updated_data %>%
mutate(climate_change_effect = case_when(
updated_data$type %in% direct_effects ~ "Direct effect",
updated_data$type %in% indirect_effects ~ "Indirect effect",
TRUE ~ "Not related"
))
#set the climate_change_effect variable to a factor
full_data$climate_change_effect <- factor(full_data$climate_change_effect)
#class(full_data$climate_change_effect)
#set a sensible order to aid plotting
full_data$climate_change_effect <- factor(full_data$climate_change_effect, levels = c("Direct effect", "Indirect effect", "Not related"))
#levels(full_data$climate_change_effect)
#check
#full_data %>% nrow()
kable(head(full_data), caption = "First 6 rows of my Dataset") %>%
kable_styling()| historic | id | classification | group | subgroup | type | subtype | iso | country | subregion | region | year | month | day | climate_change_effect |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| No | 1999-9388-DJI | nat-cli-dro-dro | Natural | Climatological | Drought | Drought | DJI | Djibouti | Sub-Saharan Africa | Africa | 2001 | 6 | NA | Indirect effect |
| No | 1999-9388-SDN | nat-cli-dro-dro | Natural | Climatological | Drought | Drought | SDN | Sudan | Northern Africa | Africa | 2000 | 1 | NA | Indirect effect |
| No | 1999-9388-SOM | nat-cli-dro-dro | Natural | Climatological | Drought | Drought | SOM | Somalia | Sub-Saharan Africa | Africa | 2000 | 1 | NA | Indirect effect |
| No | 2000-0002-AGO | nat-hyd-flo-riv | Natural | Hydrological | Flood | Riverine flood | AGO | Angola | Sub-Saharan Africa | Africa | 2000 | 1 | 8 | Direct effect |
| No | 2000-0003-BGD | nat-met-ext-col | Natural | Meteorological | Extreme temperature | Cold wave | BGD | Bangladesh | Southern Asia | Asia | 2000 | 1 | NA | Direct effect |
| No | 2000-0008-GTM | nat-geo-vol-ash | Natural | Geophysical | Volcanic activity | Ash fall | GTM | Guatemala | Latin America and the Caribbean | Americas | 2000 | 1 | 16 | Not related |
Now my data set is ready I can look at basic summaries to see if everything is expected.
#view the counts of key variables
#summary of number of disasters grouped by subgroup and type of disaster
kable(full_data %>% group_by(subgroup, type) %>% summarise(count = n()) %>% head(), caption = "First 6 rows showing counts of disasters by subgroup and type") %>%
kable_styling()| subgroup | type | count |
|---|---|---|
| Biological | Animal incident | 1 |
| Biological | Epidemic | 880 |
| Biological | Infestation | 29 |
| Climatological | Drought | 393 |
| Climatological | Glacial lake outburst flood | 3 |
| Climatological | Wildfire | 282 |
#summary of number of disasters for each region
kable(full_data %>% group_by(region) %>% summarise(count = n()) %>% head(), caption = "overall counts of disasters per continent") %>%
kable_styling()| region | count |
|---|---|
| Africa | 2032 |
| Americas | 2180 |
| Asia | 3703 |
| Europe | 1232 |
| Oceania | 358 |
#summary of number of disaster for each year
kable(full_data %>% group_by(year, region) %>% summarise(count = n()) %>% head(), caption = "First 6 rows showing counts of disaster per continent each year") %>%
kable_styling()| year | region | count |
|---|---|---|
| 2000 | Africa | 125 |
| 2000 | Americas | 101 |
| 2000 | Asia | 193 |
| 2000 | Europe | 94 |
| 2000 | Oceania | 12 |
| 2001 | Africa | 116 |
## # A tibble: 14 × 2
## type count
## <chr> <int>
## 1 Animal incident 1
## 2 Drought 393
## 3 Earthquake 626
## 4 Epidemic 880
## 5 Extreme temperature 479
## 6 Flood 3852
## 7 Glacial lake outburst flood 3
## 8 Impact 1
## 9 Infestation 29
## 10 Mass movement (dry) 13
## 11 Mass movement (wet) 423
## 12 Storm 2402
## 13 Volcanic activity 121
## 14 Wildfire 282Has there been a change in prevalence of natural disasters since 2000? Particularly has there been an increase in natural disasters related to climate change i.e. flooding?
#Overall Totals
#want a line graph which charts changes in prevalence over time split by the 3 climate conditions.
# graph where x is years, y is prevalence, split by climate change
# use summarise to create data set with total number of disaster per year not split by type or country to provide total natural disaster data
overall_disaster <-
full_data %>%
group_by(year) %>%
summarise(count = n())
plot1 <-
#plot year on the x axis, apply a stats function counting the number of rows in each condition for climate change effect
ggplot(full_data, aes(x = year, y = after_stat(count), color = climate_change_effect)) +
#add line to plot the data with the total disasters to the same graph and set the line size to 1
geom_line(data = overall_disaster, aes(y = count, color = "Total Disasters"), size = 1) +
#set the line statistic to count and line size to 1
geom_line(stat = "count", size = 1) +
#apply labels for the axis and legend
labs(x = "Year", y = "Prevalence", color = "Climate Change Effect") +
#add a title
ggtitle("Prevalence of Natural Disasters(2000 - 2022)") +
#specify colours for each line
scale_color_manual(values = c("Total Disasters" = "black",
"Direct effect" = "#E74C3C",
"Indirect effect" = "#F39C12",
"Not related" = "#616A6B")) +
#set the basic theme for the graph
theme_minimal() +
#make adjustments to the theme
theme(plot.title = element_text(size = 16), #adjust the size of the title
axis.text.y = element_text(size = 10), #adjust the size of x axis scale
axis.title.x = element_text(size = 12), #adjust the size of the x axis title
axis.title.y = element_text(size = 12), #adjust the size of the y axis title
strip.text = element_text(size = 12), #adjust the size of the facet labels
legend.text = element_text(size = 12), # adjust the size of legend text
legend.title = element_text(size = 13), # adjust the size of legend title
legend.key.size = unit(3, "mm"), # adjust the size of legend colours
panel.background = element_rect(fill = "white", colour = "white")) # make sure the background is white for saving
interactive_plot1 <-
ggplotly(plot1, tooltip = c( "y", "x")) %>% layout(width = NULL, height = NULL) #makes the plot interactive and allows the size to resize when rendered to different html screens
interactive_plot1 <-
layout(interactive_plot1, annotations = list( #add a caption to the interactive plot
text = "Data source: EM-DAT", #write the caption
x = 1.30, #set the x co-ordinate for the caption to be displayed
y = -0.05, #set the y co-ordinate for the caption to be displayed
showarrow = FALSE, #don't include an arrow
xref = "paper", #specifies the co-ordinates are respective to the whole plot
yref = "paper"
))
#save the ggplot as a png file with a white background
ggsave("output/disaster_prevelance.png", plot = plot1, bg = "white", width =6, height = 4)
#save the interactive plot as a html file
saveWidget(interactive_plot1, file = "output/interactive_disaster_prevelance.html")To use the interactive graphs hover over areas of the visualization you are interested in. If you would like to isolate conditions double click on the legend to select what you would like to see. If you would like to zoom in or out you can drag over you area of interest or use the magnifying glass icon located in the top right corner. To reset the axis press the home button in the top right corner.
There does not appear to have been a global increase in natural disasters since 2000 however the direct effect of climate change is stark. It is clear that natural disasters directly effected by climate change ( Extreme Temperature, Flood and Storm) make up the highest number of natural disaster each year. Natural disasters which are an indirect effect of climate change i.e. a consequence of natural disaster such as extreme temperature eg drought make up a smaller number of total disasters. Natural disasters not related to climate change make up a small proportion of disasters since 2000.
What regions are most effected by natural disasters?
#create a stacked bar chart with with the counts of disaster for each year. have it primarily split by climate change effect but include the types of disasters which make up each category. 1 graph per continent
#sets what the hover text shows for each data point in the interactive plot
hover_text <- paste(
"Type: ", full_data$type,
"<br>Group: ", full_data$subgroup,
"<br>Year: ", full_data$year
)
plot2 <-
#plot year on the x axis. plot the count on the y axis for each climate change category. specify that the hover_text will be shown when the graph is interactive
ggplot(full_data, aes(x = year, y = after_stat(count), fill = climate_change_effect, text = hover_text)) +
#plot the data as a stacked bar graph. set the transparency to 0.8, the line colour to white and thickness of the line to 0.1
geom_bar(position = "stack", alpha = 0.8, colour = "white", size = 0.1) +
#use facet_wrap to split the graph up by region (continent) to create 5 mini graphs. add scales = "free_x" to make sure the x axis is repeated under each graph. distribute the 5 graphs over 2 rows.
facet_wrap(~ region, scales = "free_x", nrow = 2) +
#specify the colours for each climate change category
scale_fill_manual(values = c("#E74C3C", "#F39C12", "#616A6B")) +
#add labels
labs(title = "Prevelance of Natural Disasters per Continent",
caption = "Data source: EM-DAT",
x = " ",
y = "Number of Disasters",
fill = "Effect of Climate Change") +
#apply basic theme
theme_minimal() +
#adjustments to the theme
theme(axis.text.x = element_text(angle = 45, hjust = 1, size = 10), # Add a slant on the x scale and set text size to 10
axis.text.y = element_text(size = 10), # Adjust the size of y axis scale
axis.title.x = element_text(size = 12), # Adjust the size of the x axis title
axis.title.y = element_text(size = 12), # Adjust the size of the y axis title
strip.text = element_text(size = 12), # Adjust the size of continent titles
legend.text = element_text(size = 12), # Adjust the size of legend text
legend.title = element_text(size = 13), # Adjust the size of legend title
legend.key.size = unit(3, "mm"), # Adjust the size of legend squares
plot.title = element_text(size = 16), # Adjust the size of plot title
panel.background = element_rect(fill = "white", colour = "white")) # set background to white
interactive_plot2<-
ggplotly(plot2, tooltip = c( "y", "text")) %>% layout(width = NULL, height = NULL) #make plot interactive. set the tooltip to show the y values (the count) and the hover_text information. set size of graph
# Add caption
interactive_plot2 <-
layout(interactive_plot2, annotations = list( #add a caption
text = "Data source: EM-DAT", #caption
x = 1.25, #x co-ordinate
y = -0.05, # y co-ordinate
showarrow = FALSE, #dont show an arrow
xref = "paper", #set co-ordinate to reference entire plot
yref = "paper"
))
#save ggplot as a pgn with a white background
ggsave("output/disaster_per_region.png", plot = plot2, bg = "white", width =6, height = 4)
#save interactive plot to a html file
saveWidget(interactive_plot2, file = "output/interactive_disaster_per_region.html")To use the interactive graphs hover over areas of the visualization you are interested in. If you would like to isolate conditions double click on the legend to select what you would like to see. If you would like to zoom in or out you can drag over you area of interest or use the magnifying glass icon located in the top right corner. To reset the axis press the home button in the top right corner.
When visualising how the totals of natural disasters breaks down over the continents we see the global trends repeating themselves The highest proportion of total natural disaster events in each continent are those which are a direct consequence of climate change. My second visualisation has more power to see which continent are worse effected by natural disasters. it is clear that Asia has experience a huge number of natural disasters which are both direct effects and indirect effects of climate change Oceania on the other hand has experience relatively few disasters however this may be reflective of the comparatively smaller land mass Oceania represents. The data displayed in this graph for the continent of Africa further suggests that natural disaster not effected by climate change have decreased while although it is unlikely if this is a meaningful observation.
I am pleased with the graphs I have made. I played around with lots of different graphs and ways to display my data. The dataset I originally downloaded was large with multiple interesting variables. Ultimately, I chose to focus on the variables which had to most complete data. I think adding the interactive element to the graph where viewers can hover over data points to learn more really enhanced my graph and allowed me to include more information especially on my second and main visualisation. The key information I wanted to translate was the prevalence of disasters over time split by the effect of climate change. I felt however that this was not enough and was interested in my second visualisation to not only show the split per continent but also the disasters which actually made up the climate change categories. I hope that this has translated well on different devices. An issue with my second visualisation in particular is that it can render rather small on some screens making it difficult to see events which were rare. Hopefully the interactive nature of the graph allowing you to zoom in will help.
If I had more time, I would be interested to delve deeper in to the locations of the disaster. I did attempt at first to create a world map which showed where the disaster occurred however the information for some of the small islands which make up Oceania were difficult to see on a world map. It would be interesting however to have an individual map for each continent which shows where the disasters struck and when. It would also have been interesting to look at some of the different variables on the original dataset such as cost of disaster both to lives and economy.
An issue with the EM-DAT database was firstly the incomplete data for the historical data points. This was initially disappointing as I would have loved to track the trends in natural disasters over the last century however the historical record is still being complied and will be inevitably incomplete the further back you go. Another issue I noticed was only land disasters are recorded with Tsunamis for example being absent. Even the contemporary data entries which are analysed in this project are not therefore an exhaustive list of disasters.
One of the most important skills I have learnt during this project is how to continuous notes as I am working so I always know what the code does. I have also found that keeping a record of code which doesn’t work when I am trying to figure out a problem is still helpful. After deleting code that didn’t work immediately and then regretting it later when I realised part of the code was what I needed led me to keep a code in a separate script.
Raw data available from www.emdat.be. The dataset is maintained by The Centre for Research on the Epidemiology of Disasters (CRED), UCLouvain, Brussels, Belgium.
My full repository is at https://github.com/angharad00/natural_disaster_project.